The broad, long-term, scientific objective of this project is to identify the behavioral and neuroanatomical factors that determine the efficacy of hand movement training after stroke. An important societal impact of achieving this objective will be the design of more effective robotic rehabilitation exercise technology, which will allow people with a stroke to increase their movement recovery beyond that possible with current approaches. The working hypothesis is that robot-assisted movement training following stroke is most effective if it promotes 1) effortful motor output, which produces 2) correlated, appropriate, sensations at the joints, muscles, and skin. Further, it is hypothesized that the amount of hand movement recovery that is possible with such robotic exercise is limited by 1) how much of the main outflow tract from the brain to the hand (i.e. the corticospinal tract, CST) is spared, 2) integrity of sensory processing, and 3) the cumulative history of previous movement practice. These hypotheses will be tested in a series of experiments with a robotic device that assists in grasping objects as the user works daily at home through a library of engaging video games. Aim 1 is to define the benefit of correlated sensory motor activity in robot-assisted movement training. Is it beneficial to receive robot assistance that enhances the sensations experienced during training? A robotic device will help participants to practice grasping movements that they can initiate but normally could not complete, thereby intensifying joint, muscle, and cutaneous sensation. The improvements in motor function caused by this training technique will be compared with improvements when the robot does not help with movement. Aim 2 is to define the benefit of increased motor output levels in robot-assisted movement training. Assisting movements with a robotic device as in Aim 1 can cause people to slack, decreasing their force output. By using advanced robot control software, this aim will test if training with increased relative force levels is more effective. Aim 3 is to identify the effect of patient-specific characteristics on the effectiveness of robot-assisted training. The versions of robot-assisted training to be tested in Aims 1 and 2 presume specific neural resources for optimal effect. Is it possible to predict who will benefit most from different forms of robot-assisted hand exercise? For both Aims 1 and 2, all participants' finger rehabilitation history will be measured starting within 2 weeks of stroke onset using a wearable sensor, and sensory function and the extent of CST damage will be measured when training begins, 3-6 months later. It is hypothesized that availability of CST, sensory function, and history of previous exercise will predict the response to different forms of robotic training.